Figure 1.
Illustration of Proteus phenemonen from Ioannidis and Trikalinos [27].
(Reprinted from Journal of Clinical Epidemiology, Vol. 58, J. P. Ioannidis and T. A. Trikalinos, Early extreme contradictory estimates may appear in published research: The Proteus phenomenon in molecular genetics research and randomized trials, pp. 543–549, 2005, with permission from Elsevier.) The figure shows odds ratios and 95% confidence intervals of “the relationship between the methylenetetrahydrofolate reductase (MTHFR) TT genotype in the mother and the risk of neural tube defects in the child”. The study with the strongest effect is shown by a square symbol and the study with the smallest effect is shown by a triangular symbol. The white line represents the summary odds ratio. The shaded area represents the 95% confidence interval of the summary odds ratio.
Figure 2.
Mean (thicker solid lines) and mean plus/minus one standard deviation (thinner dashed lines) of observed effects of individual studies (Eobs,pub) as a function of the number of publications.
The means and standard deviations are calculated across the 5,000 repetitions of the simulation.
Figure 3.
Mean (thicker solid lines) and mean plus/minus one standard deviation (thinner dashed lines) of the cumulative meta-analytic effect (Emeta) as a function of the number of publications.
The means and standard deviations are calculated across the 5,000 repetitions of the simulation.
Figure 4.
Mean (thicker solid lines) and mean plus/minus one standard deviation (thinner dashed lines) of the cumulative meta-analytic effect as a function of the number of studies.
Note that the number of studies can vary per repetition because the simulation was terminated when 40 publications were done under the Selective Publication Approach. Only studies having more than 4,500 out of 5,000 Emeta values available are shown (i.e., study numbers 3–585). The means and standard deviations are calculated across the repetitions of the simulation.
Figure 5.
Mean number of studies until publication as a function of the number of publications.
The means are calculated across the 5,000 repetitions of the simulation.
Figure 6.
Mean (thicker solid lines) and mean plus/minus one standard deviation (thinner dashed lines) of cumulative meta-analytic effect as a function of the number of publications.
The black lines represent the situation where Eobs is tested with respect to 0. The red lines represent the situation when ignoring the 3 latest publications for determining Emeta. The means and standard deviations are calculated across the 5,000 repetitions of the simulation.